Role-Play Large Language Models for Short Behavior Change Interventions: An Exploratory Study on Brief Action Planning

This study explores the feasibility of using Large Language Models for delivering structured behavior change interventions, focusing on Brief Action Planning for sedentary lifestyles. We leveraged a zero-shot prompting strategy without fine-tuning or providing specific data to the model. In particular, we used role-play prompting to develop an intelligent agent guided by motivational interviewing principles to support goal-setting and action-planning. The agent’s performance was evaluated through simulations and user studies, assessing its adherence to Brief Action Planning protocols. Results indicate that while role-prompting Large Language Models is a promising approach to scale up time-intensive health interventions, further research is needed to mitigate notable limitations.